fix(relations): order Ibis join_asof result deterministically (IB-REL-11)

[discreteds]

Summary

Root-caused the PR #297 CI failure: [XPASS(strict)] IB-REL-11 on test_asof_backward_strategy[ibis-duckdb], unrelated to that PR's own diff.

Root cause: DuckDB gives no row-order guarantee for ASOF JOIN output absent an explicit ORDER BY. mountainash's Ibis join_asof backend never applied one, so the test's positional assertions (result[0], result[1], ...) flipped strict-xfail/XPASS ~50/50 across independent runs — confirmed via 8 isolated process reruns both before and after the fix. Join values were always correct; only row order flapped. Same incident class as backlog item 101 (PR #293, 2026-08-16), which flagged this exact bug but did not act on it at the time.

Changes

• relsys_ib_ext_ma_util.py::join_asof — orders the Ibis asof_join result by the join key(s) before returning, matching polars/narwhals' left-order preservation for asof joins. Deterministic across 8/8 verification runs post-fix (was ~50/50 pre-fix).
• IB-REL-11 split — the ibis-duckdb portion (this bug) closed resolved_in_mountainash; the distinct, still-genuine ibis-sqlite raise (no ASOF JOIN translation in Ibis) carved out to new IB-REL-13. Updated in lockstep: divergences.py, registry/upstream-issues.yaml, docs/known-divergences.md, the test's xfail_divergence marker.
• Regenerated docs/reference/expression-coverage.{md,json} and tests/_spine_expectation_census.md (both generated files) to reflect the split.
• AGENTS.md — removed a stale 'local Ibis fork' dependency note; CI and dev both resolve ibis-framework from PyPI (>=9.0.0), no path dependency exists.

Backlog

• Filed mountainash-central item 108: a distinct, unrelated latent bug found in the same function while fixing this — join_asof's strategy parameter is silently dropped on the Ibis backend (forward/nearest behave as backward, untested anywhere). Not fixed here — needs its own per-strategy emulation design.
• Closed item 101's flagged sub-issue (item 4 in its open-questions list); its own strategic pinning/drift-detection questions (1–3) remain open.

Verification

• test_asof_backward_strategy[ibis-duckdb]: 8/8 deterministic pass, isolated process reruns (was ~5/8 flaky pre-fix).
• test_asof_backward_strategy[ibis-sqlite]: still raises UnsupportedOperationError as expected (3/3 reruns), unaffected.
• tests/relations/cross_backend/ + registry-integrity suite (test_upstream_registry_join.py, test_divergence_facts.py, test_capability_census.py): 1211 passed, 1 skipped, 26 xfailed, 0 failed.
• hatch run ruff:check on touched files: clean.
• hatch run mypy:check on the touched Ibis backend file: 5 pre-existing errors, none on touched lines (confirmed unchanged baseline).